R + Python via reticulate

Taking the radix R package or a test spin with Sckit Learn!

Objectives

1. illustrate benefit of using both Python & R, reticulate package
2. leverage R for sleek web-based reporting, new radix package
3. leverage python for machine learning algorithms and pipeline, sklearn library
4. Leverage R for excellent visualaization capabilities, ggplot2 & plotly

R Code

Take advantage of he new radix R package for sleek scientific reporting

Setup reticulate

library(reticulate)
conda_list()

             name
1     r-miniconda
2       r-gluonts
3 r-mlflow-1.16.0
4 r-mlflow-1.18.0
5    r-reticulate
6            r-tf
7       anaconda3
8    r-reticulate
                                                                                     python
1                                           /Users/gouthaman/Library/r-miniconda/bin/python
2                            /Users/gouthaman/Library/r-miniconda/envs/r-gluonts/bin/python
3                      /Users/gouthaman/Library/r-miniconda/envs/r-mlflow-1.16.0/bin/python
4                      /Users/gouthaman/Library/r-miniconda/envs/r-mlflow-1.18.0/bin/python
5                         /Users/gouthaman/Library/r-miniconda/envs/r-reticulate/bin/python
6                                 /Users/gouthaman/Library/r-miniconda/envs/r-tf/bin/python
7                                                 /Users/gouthaman/opt/anaconda3/bin/python
8 /Volumes/Data Science/Data Science Projects/bs-R-tips/renv/python/r-reticulate/bin/python

use_condaenv("anaconda3")

Switching to Python

Wine quality tutorial

https://elitedatascience.com/python-machine-learning-tutorial-scikit-learn

Setup Python modules

import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn import preprocessing
from sklearn.ensemble import RandomForestRegressor
from sklearn.pipeline import make_pipeline
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import mean_squared_error, r2_score
from sklearn.datasets import load_wine

import wine data

wine = load_wine()

data = pd.DataFrame(data = np.c_[wine["data"], wine["target"]],
                    columns= wine["feature_names"] + ["target"])

investigate data

library(tidyverse)

py$data %>% 
  as.tibble() %>% 
  glimpse()

Rows: 178
Columns: 14
$ alcohol                        <dbl> 14.23, 13.20, 13.16, 14.37, 1…
$ malic_acid                     <dbl> 1.71, 1.78, 2.36, 1.95, 2.59,…
$ ash                            <dbl> 2.43, 2.14, 2.67, 2.50, 2.87,…
$ alcalinity_of_ash              <dbl> 15.6, 11.2, 18.6, 16.8, 21.0,…
$ magnesium                      <dbl> 127, 100, 101, 113, 118, 112,…
$ total_phenols                  <dbl> 2.80, 2.65, 2.80, 3.85, 2.80,…
$ flavanoids                     <dbl> 3.06, 2.76, 3.24, 3.49, 2.69,…
$ nonflavanoid_phenols           <dbl> 0.28, 0.26, 0.30, 0.24, 0.39,…
$ proanthocyanins                <dbl> 2.29, 1.28, 2.81, 2.18, 1.82,…
$ color_intensity                <dbl> 5.64, 4.38, 5.68, 7.80, 4.32,…
$ hue                            <dbl> 1.04, 1.05, 1.03, 0.86, 1.04,…
$ `od280/od315_of_diluted_wines` <dbl> 3.92, 3.40, 3.17, 3.45, 2.93,…
$ proline                        <dbl> 1065, 1050, 1185, 1480, 735, …
$ target                         <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…

Scikit Learn-ing Time

Setup data into X (features) & Y (target) variables

y = data.target
X = data.drop("target", axis=1)

Split features into training and testing sets

X_train, X_test, y_train, y_test = train_test_split(
  X, y,
  test_size = 0.2,
  random_state = 1987,
  stratify = y
)

Preprocess by scaling X_train

scaler = preprocessing.StandardScaler().fit(X_train)

Apply transformation to X_test

X_test_scaled = scaler.transform(X_test)

Setup ML Pipeline

pipeline = make_pipeline(
  preprocessing.StandardScaler(),
  RandomForestRegressor(n_estimators = 100)
)

Setup Grid Search

hyperparameters = {
  "randomforestregressor__max_features" : ["auto", "sqrt", "log2"],
  "randomforestregressor__max_depth" : [None, 5, 3, 1]
}

Apply grid search with CV.

clf = GridSearchCV(pipeline, hyperparameters, cv = 10)
clf.fit(X_train, y_train)

GridSearchCV(cv=10,
             estimator=Pipeline(steps=[('standardscaler', StandardScaler()),
                                       ('randomforestregressor',
                                        RandomForestRegressor())]),
             param_grid={'randomforestregressor__max_depth': [None, 5, 3, 1],
                         'randomforestregressor__max_features': ['auto', 'sqrt',
                                                                 'log2']})

Get best parameters

print(clf.best_params_)

{'randomforestregressor__max_depth': None, 'randomforestregressor__max_features': 'sqrt'}

Make wine predictions & get error metrics

y_pred = clf.predict(X_test)

print(r2_score(y_test, y_pred))

0.9618984771573604

print(mean_squared_error(y_test, y_pred))

0.023166666666666665

Back to R Code

Visulaize the model with ggplot2 & plotly

Plot each predictions & actual data pair from the test set arranged by quality level to visualize how our model performing

library(tidyverse)
library(tidyquant)
library(plotly)

results_tbl <- tibble(y_pred = py$y_pred, y_test = py$y_test) %>% 
  rowid_to_column() %>% 
  arrange(y_test) %>% 
  mutate(rowid = as_factor(as.character(rowid))) %>% 
  rowid_to_column("sorted_rowid") %>% 
  pivot_longer(cols = -contains("rowid"), values_to = "value", names_to = "key")

(results_tbl %>% 
    ggplot(aes(rowid, value, color = key)) +
    geom_point(size = 0.5) +
    geom_smooth(aes(sorted_rowid, value)) +
    theme_tq() +
    scale_color_tq() +
    labs(title = "Wine Quality Level Predictions",
         x = "Row ID", y = "Quality Level")) %>% 
  ggplotly()